The repo has recently been updated to drop support for python <3.6 so that we have access to f-strings. While it may not make a huge difference here, it can make things easier to read overall.

logger.info(f"Converting {flux_var} variable to units mm/day")

It may be helpful to order these in some manner (maybe grouping by mm, cm, g, kg) just to make it easier to follow. You may want to double check with @corviday to ensure we aren't missing any cases here.

Have you considered skipping the extra variable assignment and assigning directly?

output_dataset.variables['prsn'][chunk] = prsn_data * Q_(1.0, prsn_units_from).to(prsn_units_to).magnitude

It looks like you're mixing whitespace here... the existing file uses spaces and you've inserted lines with tabs. You'll should conform to what's there already and make it consistent.

With all of the valid units that you've added to this list, it's probably worth while to eliminate some redundancy by instantiate the Unit objects in a loop. E.g. define the list as strings first and then loop over them.

valid_units = ['kg / m**2 / s' ...]
valid_units = [Unit(x) for x in valid_units]

An option that would make this more readable is to define each set of units as a variable to use in your conditional. For example:

precip_units = {Unit('kg / m**2 / s'), Unit('mm / s'), Unit('g / cm**2 / s'), Unit('cm / s')}
snow_units = {Unit('g / m**2 / s'), Unit('cm / s')}
if units in precip_units:
    ...

Do you need this conditional? It only changes one thing... the output unit in the log message. But you already have the string for the output unit string below: (units * Unit('s / day')).to_udunits_str(). Save that here, then use it both for your log message and for your attributes. Then the conditional can go away.

See "If Statements are a Code Smell".

Can you tell me more about using this as the target unit? You've changed both the mass (kg -> g) and the area (per square meter to per square centimeter). Is that definitely what we want? And is it definitely correct?

Converting from kg -> g multiplies the data by 1000 and converting from /m^2 -> /cm^2 divides the data by 10000, giving the net effect of dividing the data by 10. Since converting from mm to cm also divides by 10, I think this will work. I can check with Lee when he's back.

There's not actually any check on the data values in this test, so it's hard to know whether it's doing the conversion correctly. Could you work out what the sum snowfall in this data should be an add that to the test?

I believe the dataset used for test_process_to_prsn results in a prsn array of all zeros since these tests pass.

result = fake_dataset.variables['prsn'][:] 
result = np.where(result != 0)
for array in result:
    assert len(array) == 0

Is there possibly another dataset I could use to test the conversion?

Oh right, you had mentioned that. Well that's not a great testing dataset, then is it?! :)

Maybe you could throw a few strategic low temperature values into the test data set... it might throw some other tests out of whack, but you could adjust as necessary.

Since we're only concerned about the conversion, would it be ok if I store the pr dataset in two variables, perform the conversion on one of them, and check to make sure that each value in the two datasets differs by a factor of 10? Or would you prefer I modify the data set and tests where needed?